Sie sind auf Seite 1von 5

International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169

Volume: 5 Issue: 10 54 – 58
_______________________________________________________________________________________________

Review on Optic Disc Localization Techniques

G.Jasmine Dr. S. Ebenezer Juliet M.E., Ph.D.,


PG Scholar Department of Computer Science and Engineering
Department of computer Science and Engineering V V College of Engineering
V V College of Engineering Tisaiyanvilai
Tisaiyanvilai, India juliet_sehar@yahoo.com
jasminenesathebam@gmail.com

Abstract-The optic disc (OD) is one of the important part of the eye for detecting various diseases such as Diabetic Retinopathy and
Glaucoma. The localization of optic disc is extremely important for determining hard exudates and lesions. Diagnosis of the disease can prevent
people from vision loss. This paper analyzes various techniques which are proposed by different authors for the exact localization of optic disc to
prevent vision loss.

Keywords: Localization, Optic Disc (OD), Corner Detector, Vessel Convergence, Vessel Enhancement

__________________________________________________*****_________________________________________________

I. INTRODUCTION can lead to loss of vision. About 10,000 people lose their
vision due to diabetic retinopathy.
The retina is thin layer of tissue, in which the blood
vessels are clearly visualized. Its purpose is to receive the light Retinal Treatment is the best method for reducing the
and convert it to neural signals and these signals are send to vision loss. Regular eye examinations are required for
the brain for the visual recognition [16]. The brightest region detecting and treating the diabetic retinopathy [10].
in the retinal image is the optic disc. The blood vessels
The detection of optic disc in retinal image is very
originate from the center of optic disc. It is classified
important for the exact localization of optic disc. To find the
according to its size. The slit lamp examination is done with
abnormal structures in the image it is to be mask out from the
fundocopic lens from which vertical and horizontal disc
analysis. Optic disc detection is one of the important step for
diameter are obtained. The Optic disc cup increase in size with
diabetic retinopathy and glaucoma for screening system. The
the disc size, where large cups are obtained in healthy eyes. It
optic disc boundary and the localization of macula is used for
is also called as optic nerve head and blind spot [3]. Optic disc
the detection of exudates and diabetic maculopathy. In diabetic
is not sensitive to light because of the absence of light
maculopathy masking the false positive OD leads to improve
sensitive rods and cone in the retina, so it is called as blind
in performance of lesion detection. The OD position is used as
spot. Optic disc is located in the vessel convergence region.
a reference length for measuring the distances in retinal
The localization of optic disc is done mainly to save people
images. The location of OD becomes the starting point for
from vision loss.
vessel tracking. Thus the optic disc localization is performed
[4].

II. Techniques For Optic Disc Localization


Optic disc localization is extremely important for
determining the hard exudate lesions or nonvascularization.
By localizing the optic disc, blindness can be reduced. To
localize Optic Disc various methods are available. In this
survey some of the methods are discussed below.
2.1 Approximate nearest neighbor field based optic disc
detection
A feature match ANNF algorithm Proposed in [14],
Fig 1. Optic Disc
to find the correspondence between the optic disc images. The
correspondence between the images provides patches in the
Optic disc is mainly used for detecting various diseases query image which are close to the reference image. This
such as diabetic retinopathy and glaucoma. It can also be used method uses only one retinal image which is used to extract
for the detection of other anatomical structures such as macula the reference optic disc image. The given input image is
and retinal vessels [10]. preprocessed, in which reference optic disc is considered as an
The major health problem that has increased recently is algorithm and the target image is considered as feature match.
the Diabetic retinopathy. It is an asymptomatic disease which Then query image is preprocessed, i.e. it is equalized and
converted to grayscale. ANNF searches the nearest patches
54
IJRITCC | October 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 10 54 – 58
_______________________________________________________________________________________________
with low dimension, but it does not produce exact match. a heterogeneous database. The exudates regions are extracted
Accuracy is improved by using image coherency. The ANNF by using the morphological opening operation method. Finally
map point out the location of patches in query image which is each exudate region of 30 characteristics is extracted for the
same as of the reference image. Using these patches, the purpose of training a random forest model. A random forest
likelihood map (L) is computed. By using this L map the classifier is used to compute the risk of each exudates region.
maximum location is find out to localize the optic disc. 2.5 Fast Localization and Segmentation of Optic Disc in
2.2 Feature match: A general ANNF estimation Retinal Images Using Directional Matched Filtering and
technique and its Applications Level Sets
ANNF algorithm Proposed in [13], to estimate the The OD localization and segmentation algorithm
mapping between image pairs. This generalization is achieved proposed in [15], are used for the retinal disease screening.
by spatial-range transforms and it enables to handle various The OD locations are identified in the image using template
vision applications, 1) optic disc detection- It mainly focus on matching. It is then designed to adapt various image
medical images, for the localization of optic disc to produce as resolutions. Then the patterns in OD determine the location. A
a target image. 2) super- resolution- It focus on synthetic hybrid levelset model is used to combine both the region and
images, which uses source image. The input is a reference the local gradient information that is applied to the disc
image, which provides template for OD. It extracts the boundary segmentation. The blood vessels are removed using
template from grayscale image, after histogram equalization. morphological filtering operation. The OD size can be
The optic disc which is extracted from the image forms a calculated by FOV camera. Then the OD candidates are find
target image. The reference image is preprocessed as like out by using CIElab lightest image. Then the result is the
query image. Likelihood map is initialized, to find the optic normalized image. The optic disc candidates are selected
disc location in the image. After initialization of L map, it find according to the template responses. The segmentation uses
out the number of times each patch is mapped in the query morphological reconstruction to suppress the bright regions.
image. By using the L map it estimates the optic disc location. At last the performance is evaluated by the combination of two
2.3 Retinal vessel segmentation by improved matched test images.
filtering-evaluation on a new high-resolution fundus image 2.6 Accurate and Efficient Optic Disc Detection and
database Segmentation by a Circular Transformation
Segmentation of retinal vessels proposed in [12], to OD detection and segmentation technique proposed
segment the blood vessels in the fundus images. The in [7], is used for the circular transformation, which captures
segmentation uses MF approach in the preprocessed image. the shape of the optic disc and variation in the optic disc
The preprocessed image with five kernals, rotates into 12 boundary. First a retinal image is derived by combining both
orientations. The parametric image gets fused, so maximum the red and green components it determines the intensity of the
responses are selected for each pixel. A blood vessel tree is image. Then preprocessing takes place. According to the shape
obtained after threshold of fused image. The noise of the of the OD and the variation in the OD boundary the circular
image structures are removed using morphological operators. transformation is designed. Finally retinal image is converted
A B-spline correction method is used to improve the accuracy to the OD map. Then optic disc is localized.
of the proposed method. The two-dimensional MF exploits the 2.7 Automatic Optic Disc Detection from Retinal
correlation between image areas. The maximum response is Images by a Line Operator
selected from the set of parametric images for each pixel by An optic disc detection technique proposed in [8], is
using joint parametric image. The image is thresholded to used to extract the lightness portion in the retinal image. The
obtain vascular tree. The blood vessels are considered as retinal images are preprocessed before the detection of optic
objects and remaining parts are considered as background of disc. Then it is smoothed to enhance the brightest region
the image, where pixels inside the FOV are selected. The associated with optic disc. Then the circular regions are
result of the thresholding algorithms are evaluated and detected using the line operator. The optic disc is located
compared using HRF database. Finally, morphological accurately using the orientation of the line segment. By using
cleaning method is used to remove the artifacts, which are not orientation map the optic disc is localized. The classification is
connected to the vessel tree. done between the peak center and the surrounding of the peak
2.4 Teleophta: Machine learning and image processing center. By the combination of the peak amplitude and the
methods for teleophthalmology image intensity the optic disc is detected.
Diabetic retinopathy screening proposed in[2], is 2.8 Fast Localization of the Optic Disc Using Projection
done by using teleopthalmology screening, in which image of Image Features
deals with the normalization technique, which uses FOV Optic Disc brightness and retinal vessel orientations
photographs as size invariant. The parameters are defined features proposed in [9], based on two projections of image
based on the size. Then segmentation is done for border features that encode x coordinate and y coordinate of the optic
reflections. The reflections appear as bright and moon shaped disc. The one dimension projections determine the location of
regions. Then segmentation carried out. Then detection takes optic disc. It avoids two dimension image space and it
place and vessels are segmented and OD is detected and improves the speed of the localization process. For the
finally the macula. The vessel masks are analyzed to extract localization of OD, the process gets split into two steps. First
the structures. The detection starts with threshold to select the horizontal location of OD is determined using the image
bright regions as like optic disc. The vessel mask which is features that are projected to the horizontal axis. In Second
previously computed is used to select the correct position of step it determines the correct vertical location. Assume, if
the optic disc. An exudates detection method is developed for horizontal location is successfully identified, then the only aim
55
IJRITCC | October 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 10 54 – 58
_______________________________________________________________________________________________
is to find the correct vertical location. This can be found by maximum peak of the one dimensional signal.
feature_map_2 on the vertical axis. It is then find out as the
2.9 Automated optic disc detection in retinal images of the image is preprocessed. Then the gradient magnitude is
patients with diabetic retinopathy and risk of macular calculated by the detection of FOV in the red plane of the
edema image. The optic disc and fovea localization uses green plane.
The optic disc detection proposed in [1], based on The green plane image is blurred using a Gaussian filter to
two methodologies. On one side it belongs to the image remove the low frequency gradients. The anatomical structure
contrast and the structure filtering techniques optic disc problems are defined as a regression problem. The goal is to
location and on other side edge detection technique and Hough find the dependent variable. The threshold is applied to get a
transform are used for the circular approximation of the optic vessel map. This method combines the cues that is measured
disc. For the localization of optic disc there are three detection directly and the segmentation of retinal vasculature. The optic
methods they are, 1) Maximum difference method- which disc center is selected according to the lowest predicted
provide the difference between the maximum and minimum distance of optic disc. Using this, fovea is defined and it is
calculation of the images using filter. 2) Maximum variance selected as the fovea location in the fovea search area.
method- The maximum pixels are selected which are located 2.11Feature extraction in digital fundus images
in the brightest region. 3) Low pass filter method- It returns An exudate segmentation algorithm proposed in [5],
highest grey pixel. The OD boundary provides a circular used to enhance the exudates regions. First the image is
approximation of optic disc using segmentation. The original preprocessed to improve the contrast of the lesions. Next optic
retinography are extracted using the pixel in the optic disc. disc is eliminated using entropy. A Fuzzy c-means clustering
The coordinates are provided using localization algorithm is used to extract the exudates regions. The FCM
methodologies, 1) The blood vessels are eliminated using algorithm changes to standard FCM using spatial
dilation and smoothing are done using filter. 2) The gradient neighborhood information. Then vessel skeleton map is
magnitude is obtained using prewitt operator. 3) The noise of obtained. The neighbor pixels are obtained for each skeleton
the image is reduced using morphological erosion. 4) Finally pixel. The vessels are detected by identifying the pixels with
circular approximation of optic disc is obtained using Hough one neighbor vessel pixel. The disc boundary is detected using
transform. contour model. First image is preprocessed, then dilation
2.10 Fast detection of the optic disc and fovea in color carried out and OD boundary is determined. Then FCM
fundus photographs algorithm is used to segment the exudates. Finally SVM
KNN regressor technique proposed in [11], used to classification is used to provide the exact lesion from non-
predict the pixel distance, based on several features. The lesion.
lowest distance in the fovea is selected as fovea location. First

TABLE 1: COMPARATIVE ANALYSIS OF OPTIC DISC LOCALIZATION TECHNIQUES


PAPER TITLE ALGORITHMS/ FEATUR ES BENEFITS DRAWBACKS
METHODS
Approximate Nearest Shape, Colour, Brightness
Neighbor Field based optic  Optic Disc Location-  Performance is good.  Less accuracy.
disc detection [14] Approximate Nearest  Less computation  Dislocation of optic
Neighbour. time. disc.

Feature match: A general Colour,


ANNF estimation technique  Feature Extraction- Gradient  Accuracy is good.  Regeneration of color
and its Applications [13] Walsh Hadamard  Coherency is good. information becomes a
Transformation. difficult problem.
 K-NN Search- KD-  Differences in image
tree algorithm. intensities.

Retinal vessel segmentation Shape, Intensity


by improved matched  Preprocessing- B-  Reduces the false-  Poor Performance.
filtering-evaluation on a new Spline- based positive detections.  Evaluation is
high-resolution fundus image illumination  High Resolution. performed on low-
database [12] correction. resolution images.
 Correlation detection
between images- 2D
Match Filtering.

Teleophta: Machine learning Size, Shape, Contrast


and image processing  Preprocessing- Field  Small vessels can be  Lowers the
methods for of View. detected. specificity.
teleophthalmology [2]  Anatomical detection-  Help in rising of new  Poor Performance.
Morphological filter. population health
challenges.

56
IJRITCC | October 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 10 54 – 58
_______________________________________________________________________________________________

Fast Localization and Size, Shape, Brightness


Segmentation of Optic Disc  OD Size Estimation-  Increase robustness.  It is expensive.
in Retinal Images Using Field Of View and  Accuracy of Optic  Optic Disc detection
Directional Matched image resolution. Disc detection. is difficult.
Filtering and Level Sets  OD Segmentation-
[15] morphological
processing.
 OD Localization-
Template Matching.

Accurate and Efficient Optic Shape, Brightness


Disc Detection and  Preprocessing- Down-  Algorithm runs faster.  OD segmentation may
Segmentation by a Circular Sampling.  Accuracy gets introduce error.
Transformation [7]  OD Segmentation and improved.  Optic Disc boundary
Detection- B-Spline pixels cannot be
Filtering. determined based on
symmetry.

Automatic Optic Disc Brightness, Image Variation


Detection from Retinal  Preprocessing-  High Detection speed.  Poor Performance.
Images by a Line Operator Bilateral smoothing  Retinal lesions can be  Various types of
[8] Filter. tolerated. imaging artifacts
 Circular Detection- cannot be handled.
Line Operator.
 OD Detection- 2D
Circular convolution
Mask.

Fast Localization of the Space, Intensity


Optic Disc Using Projection  OD Localization-  Less computation  It is expensive.
of Image Features [9] Model based method. time.  Robustness for
 Search-space localization of OD is
dimensionality is difficult.
reduced.

Automated OD Shape, Colour, Depth


detection in retinal images of  OD Localization-  Accuracy is increased.  Poor Computation
patients with diabetic Maximum difference  Robustness is time.
retinopathy and risk of and variance method, increased.  High Megabytes
macular edema [1] Low-pass filter images are not
method. accepted.
 Segmentation of OD-
Dilation, Prewitt
operator, Erosion,
Circular Hough
Transform.

Fast detection of the optic Size, Space


disc and fovea in color  Preprocessing- Field  Performance is  Poor contrast.
fundus photographs [11] Of View. increased.  Does not make strong
 Position Regression –  Increase in robustness. assumption
Regression

REFERENCES
III. Conclusion
[1] A. Aquino, M. E. Gegundez, and D. Marin, “Automated optic
This paper analyzes various techniques which are proposed disc detection in retinal images of patients with diabetic
by different authors for localizing the optic disc. Diabetic retinopathy and risk of macular edema”, International Journal
retinopathy leads to vision loss, where various preventive of Biological & Life Sciences, Vol. 8, No. 11, pp. 87, 2010.
measures are taken to save people from vision loss. It focuses
[2] E. Decencire, G. Cazuguel, X. Zhang, G. Thibault, J.
on the localization of optic disc. Various approaches are used
C. Klein, F. Meyer, B. Marcotegui, G. Quellec, M. Lamard,
for OD localization they are, Approximate Nearest Neighbour
Field (ANNF), optic disc detection technique and Match R. Danno, D. Elie, P. Massin, Z. Viktor, A. Erginay, B. La,
filtering methods. These techniques are used to localize the and A. Chabouis, “Teleophta: Machine learning and image
exact optic disc. The performance is decreased due to optic disc processing methods for teleophthalmology”, fIRBMg, Vol.
dislocation. In future advanced algorithm can be used to 34, No. 2, pp. 196 – 203, 2013.
improve the performance. [3] Dehghani, Amin, Hamid Abrishami Moghaddam, and
Mohammad-Shahram Moin, “Optic disc localization in retinal

57
IJRITCC | October 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 10 54 – 58
_______________________________________________________________________________________________
images using histogram matching”, EURASIP Journal on
Image and Video Processing, No. 1, pp. 1-11, 2012.
[4] Godse, A. Deepali, and Dr. S. Dattatraya Bormane,
“Automated localization of optic disc in retinal images”,
(IJACSA) International Journal of Advanced Computer
Science and Applications 4, No. 2, 2013.
[5] G. B. Kande, P. V. Subbaiah, and T. S. Savithri, “Feature
extraction in digital fundus images”, Journal of Medical and
Biological Engineering, Vol. 9, No. 3, pp. 122–130, 2009.
[6] S. Lu, “Accurate and efficient optic disc detection and
segmentation by a circular transformation”, IEEE
Transactions on Medical Imaging, Vol. 30, No. 12, pp. 2126–
2133, 2011.
[7] S. Lu and J. H. Lim, “Automatic optic disc detection from
retinal images by a line operator”, IEEE Transactions on
Biomedical Engineering, Vol. 58, No. 1, pp. 88–94, 2011.
[8] A. Mahfouz and A. Fahmy, “Fast localization of the optic disc
using projection of image features”, IEEE Transactions on
Image Processing, Vol. 19, No. 12, pp. 3285– 3289, 2010.
[9] D. Michael, Abràmoff, K. Mona Garvin, Milan Sonka,
“Retinal Imaging and Image Analysis”, IEEE Transactions on
Medical Imaging 3, pp. 169-208, January 2010.
[10] M. Niemeijer, M. D. Abrmoff, and B. Van Ginneken, “Fast
detection of the optic disc and fovea in color fundus
photographs”, Medical Image Analysis, Vol. 13, No. 6, pp.
859 – 870, 2009.
[11] J. Odstrcilik, R. Kolar, A. Budai, J. Hornegger, J. Jan, J.
Gazarek, T. Kubena, P. Cernosek, O. Svoboda., and E.
Angelopoulou, “Retinal vessel segmentation by improved
matched filtering: evaluation on a new high-resolution fundus
image database”, Image Processing, IET, Vol. 7, No. 4, pp.
373–383, June 2013.
[12] Ramakanth, S. Avinash, and R. Venkatesh Babu,
“Featurematch: A general ANNF estimation technique and its
applications”, IEEE Transactions on Image Processing, Vol.
23, No. 5, pp. 2193–2205, May 2014.
[13] S. A. Ramakanth and R. V. Babu, “Approximate nearest
neighbor field based optic disk detection”, Computerized
Medical Imaging and Graphics, Vol. 38, No. 1, pp. 49 – 56,
2014.
[14] H. Yu, E. Barriga, C. Agurto, S. Echegaray, M. Pattichis, W.
Bauman, and P. Soliz. , “Fast localization and segmentation of
optic disk in retinal images using directional matched filtering
and level sets”, IEEE Transactions on Information
Technology in Biomedicine, Vol. 16, No. 4, pp. 644–657,
2012.
[15] http://www.healthline.com/human-body-maps/retina

58
IJRITCC | October 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________